European Export Performance

Transcription

1 European Expor Performance Angela Chepea Lionel Fonagné Soledad Zignago Absrac Using an economeric shif-share decomposiion, we explain he redisribuion of world marke shares a he level of he produc variey and by echnological conen. We decompose changes in marke shares ino srucural effecs (geographical and secoral) and a pure performance effec. We regard he EU-27 as an inegraed economy, excluding inra-eu rade. Revisiing he compeiiveness issue in such a perspecive sheds new ligh on he impac of emerging counries on he reshaping of world rade. Since 1995 he EU-27 wihsood he compeiion from emerging counries beer han he US and Japan. The EU marke shares for high-echnology producs, as well as in he upper price range of he marke, proved comparaively resilien, hough less so since he crisis. Keywords: Inernaional Trade, Expor Performance, Compeiiveness, Marke Shares, Shif-Share, European Union. JEL classificaion codes: F10, F14, F15. INRA Rennes (Angela.Chepea@rennes.inra.fr). PSE (Universié Paris 1), European Universiy Insiue, Banque de France and CEPII (lionel.fonagne@univ-paris1.fr). Banque de France (soledad.zignago@banque-france.fr).

2 1 Inroducion Emerging counries have been winning large marke shares over he las wo decades. Among hese, China sands ou wih he mos remarkable performance: i has almos rebled is world marke share since 1995, reaching 17.8% in This compeiive pressure is sriking for he mos echnological producs, where many of he new compeiors have combined an increase in marke share wih a higher uni value of he expored producs. How did EU member saes adjus o he compeiive pressure of emerging counries? Wha was he conribuion of secoral and geographical composiion of EU expors o he observed difference of performance compared o he US and Japan? Assessing compeiiveness accuraely is a challenging issue as mos of he acion is aking place on he fron of non-price compeiiveness and is poenially affeced by he producs or desinaion markes exporers specialize in. More fundamenally, he effecive demand inroduced ino macroeconomic equaions is by consrucion missing he secoral or produc dimension. Qualiy posiioning, secoral specializaion and geographical orienaion of expors all conribue o he observed changes in marke shares. Our aim in his aricle is firsly o properly measure he conribuions of produc or geographical specializaion of exporers o he observed changes in marke shares. Produclevel inernaional rade daa does no include services, we hus focus only on rade in goods. Secondly, we aim o examine how op end and high ech producs from developed counries resised o he increasing pressure of low wage exporing counries. Precisely, we develop firsly an economeric shif-share decomposiion of expor growh ha idenifies for each exporer he conribuion o he inensive margin of (i) he composiion of is expors by produc and desinaion and (ii) is compeiiveness. Accordingly, expor growh for each counry is broken down ino hree componens: a geographical composiion effec, a secoral composiion effec and an exporer effec capuring oher sources of counry s expor performance, including compeiiveness. Our second aemp, in line wih a now abundan lieraure, is o measure expor performance a he level of he (verically differeniaed) variey of raded producs (Scho 2004, Hallak 2006, Baldwin & Io 2008, Fonagné e al. 2008, Manova & Zhang 2011, Khandelwal 2010, Hallak & Scho 2011). We also evaluae 2

3 he performance of expored high-ech producs. We adop he viewpoin of an inegraed European marke and reconsruc world rade excluding inra-eu rade flows. The laer are considered as inranaional rade. 1 The mehod we use yields several improvemens wih respec o he sandard Consan Marke Share (CMS) decomposiion found in he lieraure (Tyszynski 1951, Richardson 1971a,b, Bowen & Pelzman 1984, Fagerberg 1988). 2 Firs, he economeric approach makes i possible o eliminae he non-orhogonaliy of produc and marke srucure effecs in sandard CMS analyzes, responsible for he fac ha he order of he decomposiion changes he resuls. Second, he compeiiveness effec is esimaed raher han compued as a residual of he analysis. In addiion, we are able o idenify confidence inervals for each produc, marke and exporer effec. Unlike he sandard approach, our mehodology enables us o obain resuls (effecs) ha are addiive over he ime dimension and hus ake sock of changes in counries iniial expor srucure. To proceed, i is necessary o uilize very deailed and longiudinal rade daa, covering all counries, including informaion on bilaeral rade uni values. To his end, we make use of a daabase of inernaional rade a he produc level BACI developed by Gaulier & Zignago (2010). BACI provides (FOB) reconciled values, as well as uni values (values/quaniies), of all inernaional rade flows for abou 5,000 produc headings from he 6-digi Harmonized Sysem classificaion (hereafer HS6) since We consider all raded producs, i.e. he primary and manufacuring secors, wih he excepion of mineral producs, noably oil, as well as some specific and non classified secors. The availabiliy of uni values enables us o classify flows by price range and hus o analyze he posiioning of exporers by price segmen. We employ hese daa o examine changes in marke shares of leading world exporers over he period The world disribuion of uni values for each HS6 heading allows us o classify each produc-bilaeral flow ino hree price segmens, in line wih Fonagne e al. (2008), and examine compeiion wihin op-range producs. Our daase enables also o describe changes in marke shares for high-ech producs. 1 67% of EU 27 expors are wihin he Single European Marke, where mos European counries record larger marke shares hanks o beer marke access. 2 Alernaive measures of counry compeiiveness have been used in he lieraure: comparaive advanage, specializaion or produciviy indicaors, cos of leaving indices (Fagerberg 1988, Neary 2006, Delgado e al. 2012). 3

4 In he conex of a major reshaping of world rade flows since he mid-1990s, we conclude ha he redisribuion of marke shares observed beween emerging and developed counries and among developing counries hemselves has affeced he EU, Japan and he US differenly. European marke share losses mosly concern long-sanding Member Saes. The EU s overall good performance over he period compared o he Unied Saes or Japan is associaed wih an original price-qualiy posiioning of is producs. However, his original marke posiioning of EU exporers only parially cushioned he impac of he crisis. All marke share losses recorded by he EU since 2000 were recorded over he las hree years. The res of he paper is organized as follows. We review he redisribuion of world marke shares in Secion 2, wih a focus on high-ech and op range producs. Our economeric shif share analysis of expor growh is implemened in Secion 3. Secion 4 summarizes our conclusions. 2 The redisribuion of world marke shares beween 1995 and 2010 The objecive of his secion is o ake sock of he recen shifs in world marke shares, aking ino accoun he price segmen and echnological conen of expored producs a he mos deailed available level of classificaion of raded producs. We firsly characerize he exensive and inensive margins of world rade, hen we examine wha have been he big changes in marke shares, and we conclude wih a focus on op range and high-ech producs. 2.1 Changes in rade margins Trade can increase eiher by exchanging a larger value of already raded producs beween he same parners (he inensive margin of rade), or by increasing he number of involved counries and/or exchanged producs (he exensive margin of rade). The former refers o he change in he value of exising rade flows, while he laer refers o he change in he composiion of rade flows. The enry of new compeiors is refleced in he margins of world 4

5 expors a he mos disaggregaed level of he produc classificaion. 3 Hummels & Klenow (2005) use a cross-secion of deailed rade daa o idenify he paerns of expors of 126 counries in 1995, and find ha 60% of large economies expor growh is aribuable o shipmens of a wider se of goods and he remaining 40% o larger quaniies and higher prices of each good already shipped. We adop a similar approach bu use he mos deailed rade daa compaible wih an exhausive se of exporers o compue he wo margins for he whole marix of rade flows. 4 Drawing on informaion by produc, marke, exporer, and year, we compue he exensive margin of rade, as he change in he number of rade flows a he mos deailed level, or as he ne value of appearing and disappearing rade flows. Symmerically, he inensive margin of rade is defined as he change in he value of rade flows ha are presen coninuously hroughou a given period. While a rapid urnover of rade flows can be observed in a world marix mosly full of zeros he larges conribuion o he growh in he world rade value has been on he inensive margin. To compue hese margins (Table 1), we use BACI daa from 1995 o 2010 and exclude mineral producs, specific and non-classified producs. 5. The observed USD 5,983 bn increase in world rade (column C) can be decomposed ino hree componens. Firsly, he 1.5 million elemenary bilaeral rade flows recorded in 1995 and sill in place in 2010 (second line of Table 1) have increased heir value by USD 4,773 bn. Accordingly, he inensive margin accouned for 79.8% of he change in he value of world rade (raio of column D o column C). Secondly, over one hird of 1995 rade flows (0.89 million flows) have disappeared by This is he resul of firms and counries ceasing rade wih cerain markes or cerain producs. In 1995 hese rade flows amouned o USD 324 bn. Lasly, 2.23 million new counry-parner-produc rade flows appeared during he period, corresponding 3 The exensive margin of expors so defined should no be confused wih he heerogeneous firms seings where rade inroduces a selecion beween firms, as well as, in case of muli-produc firms, a selecion wihin he porfolio of producs of each exporer. 4 Hummels & Klenow (2005) draw on HS6 daa on expors in 1995 by 110 counries o 59 imporers. Alernaively, hey use US impors from 119 counries in over 13, digi US ariff lines for he same year. Our approach also differs from Besedes & Prusa (2011) who inegrae he ime dimension ino he analysis of expor growh and breakdown he inensive margin ino a survival and a deepening componen. 5 We exclude HS chapers 25, 26, 27, 97, 98, and 99, as well as inra-eu flows, all hroughou his paper, as deailed in Secion 6.1 in he Appendix. 5

6 o he posiive exensive margin of rade. This is a very large number, comparable o he number of iniial rade flows. Overall, only 39.5% of he number of rade flows recorded in 2010 were already presen in The remaining 60.5% are new flows (column E) eiher in erms of desinaion, expored producs, or boh. Meanwhile, he conribuion of new enries o he growh of rade in value erms amouned o only 25.6%. Exis (column F) accoun for 38.0% of he number of 1995 flows bu only for 10.2% of heir value. Table 1: Exensive and inensive margins in world rade, Daa a he HS 6-digi level: Uni Inensive Exensive A B C= B-A D E F G =E-F (D+G) Enries Exis Ne Excluding USD bn 3,179 9,163 5,983 4,773 1, ,209 inra-eu rade nb flows, ,345 3,683 1,453 2, ,338 Source: Auhors calculaions using BACI values (curren USD) of raded goods. Our panel combines all rade flows, excluding inra-eu rade and mineral, specific, and non-classified producs, as well as rade wih non-independen erriories, micro-saes and small value flows (<10,000 USD). Figures are in billion dollars and in housands of HS6 bilaeral flows. The conribuion of he differen margins of rade can be compued for individual large exporers. Table 10 in he appendix compares he EU o oher large exporers from he developed and he developing world. Compuaions are performed a he counry level. For ease of presenaion, in Table 10 and in he res of he paper, resuls for counries ha accoun for less han 1% of world expors from 1995 o 2010 are aggregaed wihin hree groups Middle Eas and Norh Africa (MENA), Sub-Saharan Africa (SSA), and Res of he World (RoW). Resuls for hese hree regions and all individual counries are available in our online appendix. 6 We observe ha he conribuion of he posiive exensive margin (enries) o he growh of he value of expors is very similar for he developed economies (a mos 4%). This poins o he pronounced ineria in he expors of he advanced economies, paricularly he Japanese, German and US expors. Their rade growh is mainly accouned for by expansion in exising markes (a leas 99% for hese counries). The conribuion of 6 Zipped file a he working paper version webpage of his work. 6

7 he posiive exensive margin is conversely larger for developing and emerging economies. On average, he conribuion of new flows in expor growh for counries no repored in Table 10 is 34.4%, clearly in excess of he individual exporers repored in he Table. The lowes shares among emerging counries are observed for China and Mexico, which show a srucure of expor growh similar o he developed exporers. China confirm he magniude of he increased inensive margin, bu he diversificaion of heir expors was already accomplished in 1995 (China ships roughly as many differen producs as Germany). 7 In Secion 3 we decompose he inensive margin of expors using an economeric shifshare mehodology. 2.2 EU marke shares compared wih main world exporers In Table 2, we summarize he recen shifs in world marke shares as follows. The firs hree columns give he marke share in 1995, before he rade collapse (2007), and for he end period of our analysis (2010). In he wo subsequen columns, we repor he percenage poin changes in marke shares for he whole period and for he crisis sub-period ( ). The mos remarkable developmen in Table 2 is ha China has more han doubled is world marke share (in 2010 is marke share was 2.8 larger han in 1995), becoming a larger world rader han he US, wih 17.8% of he world marke. In 1995, EU 27 had a 20.6% marke share of he world rade in goods (excluding inra-eu flows). This marke share has been only slighly affeced by compeiive pressures from emerging economies unil he crisis, falling o 19.4% in EU counries benefied less han oher developped counries from he recovery of world rade, losing 1.4 p.p. of he world marke from 2007 o Over he enire period, Japan and he US lose around 6 percenage poins of marke share each, being more seriously affeced by he eleven-poin rise in China s share. This redisribuion of marke shares mus be gauged agains he evoluion of he euro-us dollar exchange rae. In Figure 1 we plo he evoluion of world marke shares for seleced exporers over he period , and he exchange rae agains he dollar. Despie he 7 Wang & Wei (2010) use expor a produc level for differen Chinese ciies and poin o he role of human capial and governmen inervenion in shaping a specializaion ha increasingly overlaps wih ha in high-income counries. 7

8 Table 2: Changes in world marke share for he world s larges exporers, Marke shares, %, p.p. Exporer EU France Germany Ialy Unied Kingdom Euro Area USA Japan Canada Swizerland China Brazil India Indonesia Korea Malaysia Mexico Taiwan Singapore Thailand Source: Auhors calculaions using BACI values (see foonoe of Table 1 and he daa appendix 6.1). The change in marke shares is given in percenage poins (p.p.). Resuls for counries accouning for less han 1% of world expors from 1995 o 2010 are aggregaed wihin hree groups: he Middle Eas and Norh Africa (MENA), Sub-Saharan Africa (SSA), and Res of he World (RoW). Resuls for he hree groups and all individual counries no shown here are available in he on line appendix. appreciaion of he euro, he early 2000s were a period of parial recovery for he EU s expors, wih mos of is previous losses recuperaed. Among oher indusrialized counries, Japan coninued o lose marke shares in a period of exchange rae appreciaion. The decline of he UK marke share wen in line wih an appreciaion of he pound unil 2007, wih no correcion when he exchange rae evoluion reversed. The dramaic increase in he marke share of China was accompanied by a moderae appreciaion of is currency. The appreciaion of he Brazilian real since he mid-2000s led o he sagnaion of counry s marke share around is level. Lasly, marke shares increased for India wih no connecion o he 8

9 Figure 1: Evoluions in world marke shares and exchange raes, EMU UK Japan exchr mksh exchr shbase exchr mksh China India Brazil exchr mksh exchr mksh exchr mksh Source: Auhors calculaions using BACI values (curren USD) of raded goods. Oil and inra-eu rade is excluded. BEA exchange raes agains USD. exchange rae. Finally, unil he crisis, we observe a uncorrelaed evoluion of marke shares and exchange raes, magnified for he EMU. We already noiced ha EU marke shares were severely affeced by he crisis, wih half of he losses concenraing on he year Wih his noeworhy excepion, he economic crisis has no profoundly changed he redisribuion of world marke shares among global exporers. The las column of Table 2 gives he percenage poin change in he hree-year-period , covering he grea rade collapse. The long-run rends above menioned seems o be confirmed: China s performance (+2.2 p.p. gain in world marke share beween 2007 and 2010) conrasing wih he downward rend of US marke shares (-1.2 p.p. respecively). The nex sub-secion deails he echnological dimension of larger exporers specializaion and addresses anoher dimension of inernaional compeiion: performances 9

10 differ wihin caegories of producs according o he marke posiioning of varieies. This urns o be fundamenally imporan for European exporers. 2.3 Performances in high-ech and op range producs High-ech and op range qualiy producs play an imporan role in inernaional compeiion, since hey are basically he oupu of innovaion and he source of rens. Leamer (1987) pioneered he idea ha wha one expors maers. Hausmann e al. (2007) wen one sep furher by characerizing he proximiy of specializaion beween advanced and emerging counries a he HS6 produc level. They show ha he income level of a counry s expors is a deerminan of subsequen growh. We firs focus here on high-ech producs and use he classificaion proposed by Lall (2000). Secors are classified ino primary producs, resource-based manufacures, low, medium and high-echnology manufacures, and oher ransacions. The high-ech caegory comprises elecronics and elecrical producs, as well as pharmaceuical producs, aerospace, opical and measuring insrumens, cameras, ec. (see Table 9 in he Appendix for he secors classified in he oher caegories). Counries posiioning in high-ech producs are repored in he firs wo columns of Table 3. The firs one gives he world marke shares for high-ech producs in 2010, he second one heir change in percenage poins over he period The EU has los p.p. of marke share in high-ech producs, i.e. less han he loss of p.p. for all producs (column 4 of Table 2). The Unied Saes and Japan, on he oher hand, recorded much larger losses for high ech producs han on average (respecively p.p. versus -6.4 p.p. for he US, and p.p. versus -5.8 for Japan). In he meanime, Chinese gains are very large on he high-ech marke (18.9 p.p.), due o a massive relocaion of he assembly of hese producs o mainland China. Besides some similariy in erms of expored produc caegories beween developed and developing counries (e.g. Germany and China), rade flows wih persisenly dissimilar prices can be observed wihin he mos narrowly defined producs. Though high-income and emerging economies expor quie similar bundles of goods, hey acually compee wihin indusries, on differen price-qualiy ranges (Scho 2004, 2008, Fonagné e al. 2008). Hence, 10

11 Table 3: Change in world marke shares for high-ech and op-range producs, High-ech producs Top-range Exporer % p.p. % p.p. EU France Germany Ialy Unied Kingdom Euro Area USA Japan Canada Swizerland China Brazil India Indonesia Korea Malaysia Mexico Taiwan Singapore Thailand Source: Auhors calculaions using BACI daa (see foonoe of Table 1 and he daa appendix 6.1). The change in marke shares is given in percenage poins (p.p.). Resuls for counries accouning for less han 1% of world expors from 1995 o 2010 are aggregaed wihin hree groups: he Middle Eas and Norh Africa (MENA), Sub-Saharan Africa (SSA), and Res of he World (RoW). Resuls for he hree groups and all individual counries no shown here are available in he on line appendix. specializaion occurs wihin hese caegories, on verically differeniaed varieies of producs. However, qualiy is no direcly observable. Hallak (2006) refers o produc qualiy as a demand shifer ha capures all he aribues of a produc valued by consumers. Condiional on price, a higher qualiy increases income share spen on a given variey. Using his definiion, he finds ha cross-counry variaion in uni values can be aribued o differences in qualiy. Compeiiveness ulimaely depends upon he qualiy-adjused price (Baldwin & Harrigan 2011). Baldwin & Io (2008) classify producs according o he relaed marke srucures (price compeiion versus qualiy compeiion) for nine big exporers in he pe- 11

12 riod Esimaing he price-disance relaionship separaely for each produc, hey observe more qualiy-compeiion goods in EU expors han in US and Japanese expors, and a very low share of qualiy-compeiion goods in Chinese expors. Uni values can reflec no only qualiy bu also coss (Khandelwal 2010). Idiosyncraic preferences for producs horizonal aribues may also lead o expors of goods of he same qualiy a differen prices. Finally, expor prices may vary for reasons oher han qualiy or coss (Hallak & Scho 2011). Accordingly, our approach consiss in examining changes in marke shares by price range and uses he mehod developed in Fonagne e al. (2008). If a counry s expors are in he high price range bu exhibi qualiy ha does no jusify such pricing, marke shares will shrink. Finally, we decompose each bilaeral value (X ijk ) across an addiional dimension s, corresponding o he marke segmen (s = boom, mid, op). Implemening his procedure, we observe he marke posiioning of expored producs depiced in he las wo columns of Table 3. Columns 3 and 4 give he world marke shares in 2010, and heir change in percenage poins over he period for he upper hree marke segmen. EU s leadership for op-range expors is ascerained, wih almos 27% of he world marke. The EU marke share in op range producs is 50 percen higher han per oal. The Unied Saes and Japan exhibi a quie differen paern, wih losses of respecively 30% and 40% of heir 1995 marke share for op range producs. The resilience of he EU marke share for op range producs is also remarkable, wih only 9% of is iniial marke share being los over he whole period. The evidence provided so far is purely descripive. We canno idenify he pure performance of exporing counries on his basis, as changes in marke shares can be also driven by composiion effecs. The nex secion aims o disenangle composiion effecs from pure compeiiveness. This is done for differen ranges of verically differeniaed varieies of raded producs. 12

13 3 Conribuion of compeiiveness o changes in marke shares This secion aims o idenify he conribuions o expor growh: wha are he produc and marke composiion effecs and wha sems from pure compeiiveness? One of he simples ways o invesigae growh raes is he shif-share approach, also known as he consan marke share (CMS) analysis or srucural decomposiion. Fabrican (1942) and Maddison (1952) were among he firs o formalize he shif-share decomposiion, which was exensively used aferwards. Alhough employed mainly in regional sudies on employmen and produciviy growh, his echnique has been successfully exended o inernaional rade issues over he las six decades (Tyszynski 1951, Richardson 1971a,b, Fagerberg 1988). The mehod has been exensively used in counry-level compeiiveness sudies (Laursen 1999, Wörz 2005, Brenon & Newfarmer 2007). Insead of following his radiional decomposiion, we adop an economeric approach, aking advanage of he daa disaggregaion. In addiion, in order o capure variaions across ime, we focus on he sum of annual growh in each rade flow raher han on he increase in is value beween he firs and las year of he considered period. Our mehod is herefore consrained by he observaion of he same flow in wo consecuive years (necessary for compuing annual growh raes). The 32.8 million HS6 flows ha saisfy hese condiion accoun for a rade growh of bn USD 5,653. This figure does no include rade flows creaed (bn USD 2,468) or ha disappeared (bn USD 2,137) during he period, and is larger han he inensive margin in Table 1. As previously, marke posiioning in erms of echnology or qualiy is compued from HS6 level daa. However, in order o capure even more rade flows in he inensive margin, he decomposiion of expor growhs is performed on daa aggregaed o he 2-digis level of he HS classificaion (his leads o an increase in he inensive margin from bn USD 5,653 o bn USD 5,857). 13

14 3.1 The radiional shif share decomposiion In he field of inernaional rade, he CMS or shif-share analysis aims o measure he conribuion of counries geographical and secoral specializaion o he growh of heir expors. Since he analysis is performed on expor growh, only he inensive margin of rade is explained. The radiional mehod is simply o compue he conribuion of he iniial geographical and secoral composiion of expors o changes in marke shares. The remaining proporion of he change is aribued o pure performance (i.e. price and nonprice compeiiveness). The radiional CMS is based on an algebraic decomposiion of he oal expors growh of a counry (or a region) during a given ime period. Four conribuions are idenified, namely world rade growh, growh in expors of individual producs (secoral effec), growh in impors of specific markes (geographical effec), and a residual performance of he exporer. The following equaion gives his ideniy: X i X 1 i = r X 1 i + k ( r k r ) X 1 ik + jk ( r jk r k) X 1 ijk + jk ( X ijk X 1 ijk (1+r jk) ) (1) where i denoes he exporer, j he imporer, k he produc or secor, he ime period, r he global annual growh rae of expors for all counries in he sample excep i, rk he global growh rae of produc k expors, and rjk he global growh rae of expors of produc k o counry j. Le S ik and S ijk denoe he shares of corresponding expor flows in world rade, and marke share growh raes g, gk, g jk be defined analogously o expor growh raes r, r k, r jk. When marke shares are considered insead of expor growhs, as is he case in his sudy, he CMS decomposiion has hree RHS componens raher han four. 8 S i S 1 i = k ( g k ) S 1 ik + jk ( g jk g k) S 1 ijk + jk ( S ijk S 1 ijk (1 + g jk) ) (2) The decomposiion of marke share growh raes is obained by dividing he lef and righ 8 The erm g S 1 i vanishes because g = 0 14

15 Table 4: CMS decomposiion of world marke share evoluions, , all producs: secoral effecs compued firs Change in Conribuion of: marke share Srucure effecs Performance % Secoral Geographic Addiive Mulipl. (g i ) (SECT i ) (GEO i ) (P ERF i ) (P ERF i) EU France Germany Ialy Unied Kingdom Euro Area USA Japan Canada Swizerland China Brazil India Indonesia Korea Malaysia Mexico Taiwan Singapore Thailand Source: Auhors calculaions using all rade flows exising in any wo consecuive years in he considered period (see he daa appendix 6.1). The compuaion is performed a he 2-digi level of he HS. All figures are expressed in erms of percenage change in marke share. Resuls for all individual counries no shown here are available in he on line appendix. The four columns correspond o g i, SECT i, GEO i, P ERF i, and respecively P ERF i, in equaions (4) and (5). All figures are expressed as percenages of he iniial marke share. The following ideniies hold: g i = SECT i + GEO i + P ERF i ; g i = (1 + SECT i ) (1 + GEO i ) (1 + P ERF i) 1. hand side of equaion (2) by he exporer s iniial share, S 1 i : gi = S i S 1 i S 1 i = k ( ) S 1 ik gk S 1 + i jk ( ) S 1 g jk gk ijk S 1 + i jk ( S ijk S 1 i S 1 ijk S 1 i (1 + g jk) ) (3) The hree erms of equaion (3) correspond respecively o he he secoral srucure, ge- 15

16 ographic srucure and expor compeiiveness (performance) effecs and are compued for each exporing counry i. The decomposiion for he enire period is obained as he produc of annual srucural and performance effecs: 9 g i = [1 + g i] 1 = SECT i + GEO i + P ERF i (4) A correcion erm e i is needed o ensure he equaliy of lef and righ hand side expressions on equaion (4). 10 We assimilae his erm o he performance effec P ERF i. This is in accordance wih he convenion of he radiional CMS approach of compuing expor performance as he residual growh rae afer deducing he conribuion of srucural effecs. Alernaively, he marke share evoluion over he enire period can be decomposed ino a produc of hree erms: g i = (1 + SECT i ) (1 + GEO i ) (1 + P ERF i) 1 (5) In his case counry s expor performance is compued slighly differenly, 11 and includes again a correcion erm v i. Such srucural decomposiion has a major drawback: resuls are sensiive o he order in which he composiion effecs are considered. Compuing secoral effecs firs and geographical effecs aferwards and vice versa yields differen resuls. Reversing he order of secoral and geographic srucure erms in equaion (2) yields differen amouns for he wo effecs, bu leaves almos unalered he expor compeiiveness residual: S i S 1 i = j ( gj ) S 1 ij + jk ( g jk g j) S 1 ijk + jk ( S ijk S 1 ijk (1 + g jk) ) (6) 9 SECT i = [ 1 + k [ P ERF i = 1 + jk 10 See foonoe 9. [ 11 P ERF i = (g k ) ( S 1 ik (( 1 + jk S ijk /S 1 i (( /S 1 i ) S ijk /S 1 i ) ], GEO i = 1 + jk ( ) ) ] S 1 ijk /S 1 i (1 + gjk ) + e i 1. ) ( ) ) ] S 1 ijk /S 1 i (1 + gjk ) v i. [ ( ) ( ) ] gjk g k S 1 ijk /S 1 i, 16

17 Table 5: CMS decomposiion of world marke share evoluions, , all producs: geographic effecs compued firs Change in Conribuion of: marke share Srucure effecs Performance % Geographic Secoral Addiive Mulipl. (g i ) (GEO i ) (SECT i ) (P ERF i ) (P ERF i) EU France Germany Ialy Unied Kingdom Euro Area USA Japan Canada Swizerland China Brazil India Indonesia Korea Malaysia Mexico Taiwan Singapore Thailand Source: Auhors calculaions using all rade flows exising in any wo consecuive years in he considered period (see he daa appendix 6.1). The compuaion is performed a he 2-digi level of he HS. Resuls for all individual counries no shown here are available in he on line appendix. All figures are expressed in erms of percenage change in marke share. The four columns correspond o g i, GEO i, SECT i, P ERF i, and respecively P ERF i, in equaions (4) and (5). All figures are expressed as percenages of he iniial marke share. The following ideniies hold: g i = GEO i + SECT i + P ERF i ; g i = (1 + GEO i ) (1 + SECT i ) (1 + P ERF i) 1. Table 4 displays resuls of he radiional CMS decomposiion, wih secoral srucure effecs compued firs. Calculaions were performed a he 2-digi level of he HS, on he inensive margin only: we use all rade flows exising in any wo consecuive years in he considered period. Flows associaed wih HS secions 25, 26, 27, 97, 98, 99, iny values (below USD 10,000), non-independen erriories and micro-saes were excluded. In he 17

18 firs column we show he percen change in he marke share compued on he inensive margin over he period for our seleced sample of counries (e.g % for he EU). The subsequen columns show he performance effec compued alernaively using an addiive decomposiion for he enire period as in Equaion (4) and for he muliplicaive approach in Equaion (5). The las wo columns show he geographic and secoral effecs when he role of he secoral composiion is compued firs. The secoral effec is posiive (+3.4%) for he EU, bu much weaker han he geographic effec (resp. +6.2%). Table 5 display resuls of he radiional CMS decomposiion, bu wih geographic srucure effecs compued firs. The hree firs columns are indeed idenical o he firs columns of Table 4. However, we obain very differen magniudes for geographic and secoral effecs. In he EU case, he secoral effec is now larger han he geographic effec, as opposed o he former decomposiion. Even he sign of he effecs for a given counry can be reversed. For he US, he secoral effec is posiive when he secoral effecs are compued firs, bu becomes negaive when geographic effecs are compuer firs. 3.2 The economeric shif-share approach Considering he shorcomings of he radiional mehod, we rely on an alernaive shif-share mehodology, based on economerics, proposed by Chepea e al. (2005), which is a furher developmen of he weighed variance analysis of growh raes of Jaye (1993). 12 Similarly o he radiional CMS, he aim of his mehod is o ulimaely decompose he growh of each counry s world marke shares ino hree erms: a geographical srucure effec, a secoral effec, and an exporer-effec which represens he exporer s performance. Bu conrary o he radiional approach, i relies on economerics raher han on simple algebra. compue counry-level srucural and performance effecs, we firs explain he growh rae of each individual rade flow (from each exporer o each imporer for a given produc and year) and, in a second sep we aggregae resuls a he exporer level. 1 2 Le w denoe he average weigh of a flow in world rade in years 1 and : w ( ) ijk ( = X 1 ijk + X ijk and w X 1 X i = 1 X 1 i + X 2 X 1 i ). The bilaeral and secoral expor growh raes are X 12 The radiional shif-share analysis is acually a consrained and imperfec version of regression and variance analysis echniques. 18 To

19 regressed on dummies idenifying exporers (i), imporers (j) and HS2 groups of producs (k) wih weighed (by w ijk ) OLS: ln ( ) X ijk X 1 ijk where X represens he value of expors, β j and γ k = α i + β j + γ k + ε ijk. (7) capure he conribuion of he average global geographical and produc rade srucure in year o he annual growh rae of expors beween 1 and, and α i is he amoun of growh in ha can be aribued o he expor performance of counry i. One of he advanages of he economeric shif-share approach, wih respec o he radiional approach, is he esimaion of sandard errors for each effec, which can be used o predic he saisical significance of counry-level expor performance and srucural effecs. We sugges wo mehods for compuing sandard deviaions for each erm of he above decomposiion as deailed in Secion 6.3. We decided o rely on he Dela mehod providing wih more accurae sandard errors, hough imposing more compuaional consrains. The shif-share decomposiion is performed for each year beween 1995 and We hus esimae fifeen annual effecs for each exporer, imporer and produc. 13 Unlike Chepea e al. (2005), he growh rae of counry i s expors is compued here as he logarihm of he Törnqvis index of is expors of each produc k o each parner j. 14 The annual growh of counry i s expors in period is obained as an approximaion of he rue logarihmic change in is expors: ( ) X d ln Xi = ln i X 1 i jk wijk wi ln ( ) X ijk X 1. (8) ijk Thus, we express he growh of counry i s expors as a weighed average of he logarihmic change in is expors of each produc k o each parner j Equaion (7) is esimaed for , ,..., growh raes. Daa on 1994 flows are used o compue weighs wij 1995 and wi 1995 and as reference for obaining resuls in volume erms (Table 11 of he Appendix). 14 The Törnqvis index is he weighed geomeric average of he relaive change beween he curren and base period where weighs are he arihmeic average of he marke shares in he wo periods. 15 Alhough a he exporer/imporer/produc level he difference beween growh raes compued according o he wo sides of he above equaion may vary significanly, he weighed averages a he level of each 19

20 Combining equaions (7) and (8), we can express he overall growh of counry i expors in erms of he hree ypes of effecs menioned above: d ln X i = α i + j wij wi β j + k w ik w i γ k. (9) To reach equaion (9) we use he fac ha he weighs of all flows involving exporing counry i add up o he weigh of is expors in world rade, wi = jk w ijk, and ha he sample weighed average of he error erm in (7) is equal o zero, jk w ijk ε ijk = 0.16 Given he large size of our sample (over 200,000 observaions per year), he ideniy esablished by (9) is almos unalered if we replace he consan erm, exporer, imporer, and produc effecs by heir OLS esimaes. Le has indicae OLS-esimaed coefficiens in (7). When esimaing (7), one imporer and one produc fixed effecs is removed because of collineariy. 17 Therefore, ˆα i is a measure of counry i s pure expor growh relaive o he omied parner counry and raded produc. A measure of counry i s effec independen of he choice of he omied counry is given by he leas square mean, obained by adding he inercep and he weighed mean of parner and produc effecs o he esimaed effec: α i = ˆα i + j w j ˆβ j + k w k ˆγ k. (10) Noe, ha he weighed average of counry-specific pure expor growh gives he growh rae of world rade: i w i α i = ( ) ijk w ijk ln Xijk = d ln X. We employ he fac ha X 1 ijk ( he sum of weighs across any dimension is equal o one i w i = j w j = ) k w k = 1 o esablish his resul. For similar reasons, we normalise he esimaed imporer and produc effecs. The new values are obained by subracing he weighed average of esimaed effecs from he parameers esimaed originally: β j = ˆβ j j w ˆβ j j and γ k = ˆγ k k w kˆγ k. Noe ha wih exporer are very similar. For example for France he difference beween he wo weighed means represens a mos 6% of he larges of he wo values. For Germany he difference is even smaller. 16 The las consrain is implicily imposed when esimaing (7) wih weighed OLS. 17 Dropping he consan permis o keep all exporer fixed-effecs. 20

21 ( ) Xijk hese noaions equaion (7) becomes ln = α X 1 i + β j + γ k + ε ijk. The decomposiion ijk (9) can hen be re-wrien as: d ln X i = α i + j wij wi β j + k w ik w i γ k. (11) The firs righ-hand side elemen of (11) represens he expor performance of counry i. The las wo erms reflec he conribuion of is expors srucure by parner and produc o he overall growh of is expors. We refer o hem as he geographical and secoral srucure effecs. The decomposiion of expor growh is carried ou separaely for each year. Noe ha he sum of annual growh raes yields he change in he value of expors beween he firs and las year of he period. Therefore, resuls for he enire period are obained by adding ogeher he differen effecs across years: d ln X i d ln X i = α i + ( j wij wi β j ) + ( k w ik w i γ k ). (12) Now, we can ranspose his decomposiion ino a decomposiion of changes in marke shares. For his, we subrac from boh he lef and righ-hand side expressions of (12) he logarihmic change in world expors over he period, compued as a Törnqvis index, d ln X 95 10, and ake he exponenials of he resuling expressions. 18 We obain: g i exp ( d ln X i d ln X 95 10) 1 = (1 + Perf i ) (1 + Geo i ) (1 + Sec i ) 1 (13) where Perf i = exp ( α i d ln X ) 1, and Geo i and Sec i are he exponenials of he las wo erms of he righ-hand side expression of equaion (12) minus one. Noe ha d ln X i and d ln X are approximaions of rue logarihmic changes in counry and world expors obained wih he Törnqvis index. 19 Therefore, g i in equaion (13) 18 Accordingly, we have d ln X (d ln X ) = ( ) wi d ln X i. 19 i d ln X i ln ( ) Xi 2007 /Xi 1995 and d ln X ln ( X 2007 /X 1995). 21

22 is an approximaion of he acual marke share growh rae. 20 Exporing counries have no influence on srucural effecs affecing heir expors. These effecs resul from he growh in desinaion markes, given he geographical and secoral composiion of expors. In conras, he performance effec is a rue compeiiveness effec. I indicaes he degree o which he exporing counry has been able o gain or lose marke shares, afer conrolling for composiion effecs. 3.3 Economeric decomposiion of changes in world marke shares: all producs We now repor he resuls of he shif-share analysis. We explain he annual growh of all rade flows exising in any wo consecuive years and aggregae resuls in erms of marke shares over he period The esimaion is performed a he 2-digi level of he HS: he 6-digi level does no give very differen resuls, while he HS2 secures higher saisical significance of parameer esimaes. However we coninue o define uni values ranges and echnological producs a he HS6 level. The saisical significance of fixed effecs is indicaed in he ables of resuls (see Appendix for deails on heir compuaion). Table 6 shows he differences beween marke shares considered in his secion and hose in Secion 2. The firs column in Table 6 repors he changes in marke shares beween 1995 and 2010 for boh he inensive and exensive margin, as presened in Table 2 (e.g. he EU25 loses 2.6 p.p. of he world marke shares). These amouns are compued as: Xi 2010 X 100 X1995 i 100 (14) 2010 X1995 The following hree columns consider he change in world marke shares by focusing on he inensive margins of rade only. Column (2) gives changes in original marke shares aribued o he inensive margin compued on an annual basis (i.e. for flows exising in any wo consecuive years). I is obained by excluding he exensive margin from he firs ( ) ( 20 X 2007 i Acual (rue) marke share growh raes are obained as X X i X 1995 X / 1995 i X ) As menioned above and in he daa appendix, we eliminae from our sample he noise associaed wih very small values (below USD 10,000), non-independen erriories and micro-saes, and drop HS secions 25, 26, 27, 97, 98, 99 (mineral, specific and non-classified producs). 22

23 numeraor: where 2010 X i = Xi =1995 jk X i 2010 X ( X +1 ijk 100 X1995 i 100 (15) 2010 X1995 X ijk ) I [ X ijk 0 X +1 ijk 0] and I [.] is an indicaor variable (dummy) aking he value one when he condiion under he brackes is verified and zero oherwise. Noe ha he difference beween columns (1) and (2) is negligible for all counries. This indicaes ha he change in marke shares for he inensive margin is a good proxy of he change in marke shares compued from all rade flows. Column (3) provides marke share evoluions compued on he inensive margin only, i.e. using he exac sample on which we perform he shif-share analysis. Unlike in column (2), he global marke in each year is defined as he sum of posiive rade flows ha do no vanish by he nex year. Accordingly, evoluions in column (3) are obained as follows: X i X i (16) X 2010 X 1995 where X 1995 i = jk ( X 1995 ijk I [ X 1995 ijk 0 X 1996 ijk 0 ]) and X i = X i Xijk 1995 I [ Xijk 1996 = 0 ]. 22 jk Global rade corresponding o he inensive margin is simply he sum of rade values compued for each exporing counry. 23 Column (4) displays he Törnqvis approximaion of changes in column (3), expressed in percenage erms of exporers iniial shares, i.e. g i from equaion (13) where X ṭ are replaced wih X ṭ. The decomposiion of he change in column (4) according o our economeric shif-share analysis is repored in he las hree columns of Table 6. To clarify he difference beween he differen columns of Table 6, le us consider he case of Chinese expors. In 1995 Chinese expors represened only 6.3% of he value of world rade; by he year 2010 hey increased by p.p. When we exclude he exensive margin (flows ha appeared and disappeared over he period), he marke share growh is almos unchanged: p.p. or p.p., depending on how one defines he global marke 22 X2010 i = 23 X = i 1995 X i =1995 jk X i. ( ) [ X +1 ijk X ijk I Xijk 0 X+1 ijk ]. 0 23

24 Table 6: Changes in world marke shares for large exporers, : overall growh, inensive margin and shif-share decomposiion Overall Inensive margin Shif-share as in he global marke defined as Expor Srucural effecs Table 2 Σ all flows Σ surviving flows performance geographical secoral p.p. p.p. p.p. % % % % Exporer (1) (2) (3) (4) (5) (6) (7) EU France Germany Ialy UK Euro Area USA Japan Canada Swizerland China Brazil India Indonesia Korea Malaysia Mexico Taiwan Singapore Thailand Source: Auhors calculaions using BACI daabase (see he daa appendix 6.1). Figures in column (1) are obained using he sample of he panel (1) of Table 1. Column (2) is obained by excluding he exensive margin from he numeraor. Column (3) provides he same informaion as column (2), bu compued on he inensive margin only. This value is approximaed by he Törnqvis index in column (4). Surviving flows for each year are flows ha exis no only in, bu also in 1 and/or + 1. The shif-share esimaion is performed a he 2-digi level of he HS. The las four columns correspond o erms g i, P erf i, Geo i and respecively Sec i (g i = (1+P erf i ) (1+Geo i ) (1+Sec i ) 1, equaion 13), in percenage form. Recall ha he order of decomposiion does no maer wih his mehod. ***, **, ** denoe significance a he 1%, 5% and 10% level. Resuls for counries accouning for less han 1% of world expors from 1995 o 2010 are aggregaed wihin hree groups: he Middle Eas and Norh Africa (MENA), Sub-Saharan Africa (SSA), and Res of he World (RoW). Resuls for he hree groups and all individual counries no shown here are available in he on line appendix. 24

25 (he sum of all flows or only of flows ha exis as well in he previous and/or following year). The p.p. gain represens a 192.3% increase in he iniial Chinese world marke share (column 4). 24 The 192.3% growh in expors from China is he sum of he secoral, geographical and performance effecs, compued for each year of he period, here in a muliplicaive form [( ) ( ) ( ) 1]. In he EU27 case, he 12.5% marke share loss according o he Törnqvis approximaion (column 4, Table 6) is decomposed in columns (5) o (7). This loss, driven by he negaive performance effec, was smoohed by he geographical and secoral effecs. The secoral effec conribued for more han half of his smoohing. However, he magniude of he EU s losses is much more limied han hose recorded by Japan and he US (resp % and -34.9%). All in all, he EU s performance compares favorably wih he US or Japan given he pressure of new compeiors. China, bu also India and Indonesia show impressive expor performances, alhough experience negaive secoral conribuions in general. 25 Mexico is penalized by he geographical srucure of is expors. The resilience of EU s marke shares is largely due o Germany s resilience and, o a lesser exen, o new Member Saes performances. 26 Moreover, he EU s losses are smaller in volume erms (Table 11 in he Appendix), indicaing a negaive price effec, in paricular for Germany and France. 3.4 Comparison beween he wo shif-share mehods We now compare effecs obained wih he radiional CMS analysis (boh wih addiive and muliplicaive erms) and wih he economeric approach. Marke share growh raes 24 China s inensive margin accouned for p.p. of he overall global marke (including boh flows ha survive and vanish from year o year), and for p.p. of he global marke compued on he inensive margin (as he sum of flows ha exis as well in he previous and/or following year). These changes represen, respecively, 180.0% and 188.5% of he corresponding global markes. When expor growh raes are compued wih he Törnqvis approximaion, he 188.5% Chinese marke share growh becomes 192.3%. 25 As confirmed by Belramello e al. (2012) using our mehodology and daa, he secoral effec is negaive for mos emerging exporers, reflecing heir specializaion oward more radiional, lower echnology indusries. The CMS analysis from Crespo & Fonoura (2010), which uses a panel similar o ours, also provides evidence of he growh of marke share of many emerging counries in Asia and Cenral and Easern Europe, despie heir negaive secor and /or geographical srucure effecs. 26 Deailed resuls by individual EU27 counries are available in he online Appendix. 25